fix(voice): prefer requested MP3 for playback

cli.py

@@ -12005,11 +12005,14 @@ class HermesCLI(CLIAgentSetupMixin, CLICommandsMixin, CLIBillingMixin):
                 tts_result = json.loads(raw_result) if isinstance(raw_result, str) else {}
             except Exception:
                 tts_result = {}
-            audio_path = tts_result.get("file_path") or mp3_path
 
-            # Play the actual file returned by the TTS provider. Command
-            # providers may keep native formats such as FLAC/WAV instead of
-            # writing the requested MP3 path.
+            # Prefer the requested MP3 when the provider produced it. This
+            # preserves reliable local playback while still supporting
+            # providers that write to and return a different path.
+            audio_path = mp3_path
+            if not os.path.isfile(mp3_path) or os.path.getsize(mp3_path) == 0:
+                audio_path = tts_result.get("file_path") or mp3_path
+
             if os.path.isfile(audio_path) and os.path.getsize(audio_path) > 0:
                 play_audio_file(audio_path)
                 # Clean up

hermes_cli/voice.py

@@ -854,7 +854,13 @@ def speak_text(text: str) -> None:
             tts_result = json.loads(raw_result) if isinstance(raw_result, str) else {}
         except Exception:
             tts_result = {}
-        audio_path = tts_result.get("file_path") or mp3_path
+
+        # Prefer the requested MP3 when the provider produced it. This
+        # preserves reliable local playback while still supporting providers
+        # that write to and return a different path.
+        audio_path = mp3_path
+        if not os.path.isfile(mp3_path) or os.path.getsize(mp3_path) == 0:
+            audio_path = tts_result.get("file_path") or mp3_path
 
         if os.path.isfile(audio_path) and os.path.getsize(audio_path) > 0:
             _debug(f"speak_text: playing {audio_path} ({os.path.getsize(audio_path)} bytes)")

tests/hermes_cli/test_voice_wrapper.py

@@ -294,20 +294,44 @@ class TestSpeakTextGuards:
         from tools import tts_tool
 
         played = []
+        returned_path = "/tmp/hermes_voice/actual.flac"
 
         monkeypatch.setattr(
             tts_tool,
             "text_to_speech_tool",
-            lambda **_kwargs: '{"success": true, "file_path": "/tmp/hermes_voice/actual.flac"}',
+            lambda **_kwargs: f'{{"success": true, "file_path": "{returned_path}"}}',
         )
         monkeypatch.setattr(voice.os, "makedirs", lambda *_args, **_kwargs: None)
+        monkeypatch.setattr(voice.os.path, "isfile", lambda path: path == returned_path)
+        monkeypatch.setattr(voice.os.path, "getsize", lambda _path: 1000)
+        monkeypatch.setattr(voice.os, "unlink", lambda _path: None)
+        monkeypatch.setattr(voice, "play_audio_file", lambda path: played.append(path))
+
+        assert voice.speak_text("Hello world") is None
+        assert played == [returned_path]
+
+    def test_speak_text_prefers_requested_mp3_over_returned_ogg(self, monkeypatch):
+        import hermes_cli.voice as voice
+        from tools import tts_tool
+
+        played = []
+        requested_paths = []
+
+        def fake_tts(**kwargs):
+            requested_path = kwargs["output_path"]
+            requested_paths.append(requested_path)
+            ogg_path = requested_path.rsplit(".", 1)[0] + ".ogg"
+            return f'{{"success": true, "file_path": "{ogg_path}"}}'
+
+        monkeypatch.setattr(tts_tool, "text_to_speech_tool", fake_tts)
+        monkeypatch.setattr(voice.os, "makedirs", lambda *_args, **_kwargs: None)
         monkeypatch.setattr(voice.os.path, "isfile", lambda _path: True)
         monkeypatch.setattr(voice.os.path, "getsize", lambda _path: 1000)
         monkeypatch.setattr(voice.os, "unlink", lambda _path: None)
         monkeypatch.setattr(voice, "play_audio_file", lambda path: played.append(path))
 
         assert voice.speak_text("Hello world") is None
-        assert played == ["/tmp/hermes_voice/actual.flac"]
+        assert played == requested_paths
 
 
 class TestContinuousAPI:

tests/tools/test_voice_cli_integration.py

@@ -1128,10 +1128,34 @@ class TestVoiceSpeakResponseReal:
     def test_play_audio_uses_returned_tts_file_path(
         self, _tts, mock_play, _mkd, _isf, _gsz, _unl, _cp
     ):
+        _isf.side_effect = lambda path: path == "/tmp/hermes_voice/actual.flac"
         cli = _make_voice_cli(_voice_tts=True)
         cli._voice_speak_response("Hello world")
         mock_play.assert_called_once_with("/tmp/hermes_voice/actual.flac")
 
+    @patch("cli._cprint")
+    @patch("cli.os.unlink")
+    @patch("cli.os.path.getsize", return_value=1000)
+    @patch("cli.os.path.isfile", return_value=True)
+    @patch("cli.os.makedirs")
+    @patch("tools.voice_mode.play_audio_file")
+    @patch("tools.tts_tool.text_to_speech_tool")
+    def test_play_audio_prefers_requested_mp3_over_returned_ogg(
+        self, mock_tts, mock_play, _mkd, _isf, _gsz, _unl, _cp
+    ):
+        def fake_tts(**kwargs):
+            mp3_path = kwargs["output_path"]
+            ogg_path = mp3_path.rsplit(".", 1)[0] + ".ogg"
+            return f'{{"success": true, "file_path": "{ogg_path}"}}'
+
+        mock_tts.side_effect = fake_tts
+
+        cli = _make_voice_cli(_voice_tts=True)
+        cli._voice_speak_response("Hello world")
+
+        requested_path = mock_tts.call_args.kwargs["output_path"]
+        mock_play.assert_called_once_with(requested_path)
+
 
 class TestVoiceStopAndTranscribeReal:
     """Tests _voice_stop_and_transcribe with real CLI instance."""